Carbon rich materials lacking sufficient oxygen to undergo complete combustion have long been known to produce nanocarbon condensates of utility to industries spanning nanomedicine to quantum computing, when subject to strong shockwaves. However, the associated extreme conditions (e.g. 1000s of K and 10s of GPa) and rapid system evolution (e.g. 10s of ps) has precluded a clear understanding of early time phenomena giving way to carbon condensate formation. The semi-quantum density functional theory tight binding (DFTB) simulation method is ideal for studying chemistry on these timescales, offering much of the predictive power of density functional theory (DFT) at a fraction of the computational cost. However available parameterizations are not designed for application to organic molecular materials under extreme conditions.
Here, we describe a new machine learning (ML) approach for rapidly tuning DFTB models to simulate molecular materials under extreme conditions and demonstrate its application to modeling of 3,4-bis(3-nitrofurazan-4-yl)furoxan (i.e. DNTF), which has recently been shown to produce liquid carbon nanodroplets upon detonation that subsequently solidify into graphitic nano-onions. We investigate early shockwave-driven decomposition chemistry to determine (1) major chemical kinetics steps, (2) the DNTF shock equation of state, and (3) implications of (1) and (2) for the DNTF nanocarbon formation mechanisms. We find evolution to be characterized by release of CO2, N2, and CO, as well as large CxNyOz species that are likely to be precursors to the experimentally observed carbon nano onions. Moreover, we find O purification (i.e. via CO2 elimination) more rapid than that of N (i.e. via N2 elimination), consistent with the experimentally observed N-containing species entrainment within the carbon condensates. Ultimately, we find the present developed ML-driven DFTB tuning approach well suited to the study of chemistry under extreme conditions, by providing a means of achieving long-timescale simulation with DFT-accuracy.